Linux kernel 2.6.23 released with a wee bit of controversy

Another update of the Linux kernel slipped out this week with a new process scheduler, virtualization options and a wee bit of controversy to boot. Linux 2.6.23, which was released October 9, incorporates a new process scheduler called the Completely Fair Scheduler that gives more CPU horsepower to applications that are more user interactive but that also ensures (through a "fairness" engine) reasonable CPU resources are allocated to ordinary running tasks.

One idea for balancing the needs of newer interactive workloads and ordinary processes came from a developer named Con Kolivas, who fine tuned the scheduler algorithm in a way that treated all processes fairly. He included it in a process scheduler called RSDL. In the end, the kernel project went with an even newer process scheduler, CFS, which was developed by another developer Ingo Molnar, which does give priority to more interactive processes but also incorporates Kolivas' fairness doctrine.

This choice generated some amount of controversy, especially in light of Con's subsequent announcement that he would quit the kernel development process. The hoopla has since died down after Con publicly denied his departure had anything to do with the decision to use CFS. The decision was based on technical merit, and nothing else, the kernel group maintains. It was an interesting addendum, though, to the intense focus on developer rights spurred by the Sun-OpenOffice controversy over the last week.

It wasn't the only aspect of the kernel update that generated some noise.

The Linux 2.6.23 kernel also offers a new hypervisor for Linux guests (only) on Linux dubbed lguest. The Linux kernel update also offers KVM guest support and merges part -- but not all -- of Xen.

The newest lguest hypervisor is a simple, 5000 lines of code written by Rusty Russell. It uses the paravirtu_ops infrastructure but unlike KVM, it does not require VT/SVM hardware.

Also according to kernelnewbies.org, Russell encourages developers to "fork" lguest to create a better hypervisor for the Linux kernel. And there's better support for Xen, notably the ability for the kernel to boot in a para virtualized environment using the Xen hypervisor.

But support for the hypervisor is not included. Support is limited to Xen-based guests only.

Choice, and continued innovation, are good. Nevertheless, some in the community are speculating that the dizzying array of virtualization engines coming out of the open source community -- and no standard, widely embraced virtualization engine for the Linux platform -- could paralyze corporate decision making and stymie growth of Linux. Another interesting addendum in a week fraught with concerns about forking and fragmentation.